[hansent]

What changed and why

roboflow_core/rate_limiter@v1 is meant to fall back to datetime.now() when video_reference_image is not a video frame. It never did: WorkflowImageData.video_metadata doesn't raise for an image without metadata. Instead it returns a placeholder with frame_number=0, fps=30, so the try/except never fell back. The limiter computed time 0 on every run and, after the first run, suppressed next_steps forever. This affects still images (HTTP requests) and images from steps that build a new WorkflowImageData without copying video metadata (stitch_images@v1, grid_visualization@v1, stability_ai_*). It is the follow-up split out of #3296, where Camera Calibration dropping metadata exposed it.

• Add WorkflowImageData.has_video_metadata(), a public check for real (producer-supplied) video metadata, so blocks don't read _video_metadata.
• The limiter now uses:
  • wall-clock time when there is no reference image or it has no video metadata;
  • video time (frame_number / fps) only for comes_from_video_file=True with a non-zero FPS;
  • frame_timestamp otherwise (streams, unknown source, FPS missing or 0). This is the same rule Time in Zone and Velocity use.
• Keep the last-execution time as float seconds, so naive wall-clock and timezone-aware frame timestamps never get subtracted from each other.
• Update the block's docs and field descriptions.

Behavior change for streams: frames with comes_from_video_file set to False/None that report an FPS used to be timed by frame_number / fps. They now use frame_timestamp. Video files are unchanged.

Execution engine changelog: not needed. has_video_metadata() is an additive accessor and changes no compile or run behavior (see .cursor/rules/execution-engine-version-changelog.mdc). The fix is recorded under ## Unreleased → ### Fixed in workflows/CHANGELOG.md.

Tests

• New multi-frame unit tests in workflows/tests/unit_tests/core_steps/control_flow/test_rate_limiter.py, with wall-clock time simulated by patching the module's datetime:
  • an image without metadata, and no reference image (wall-clock path);
  • a video file at 1 FPS, frames 1, 3, 5 (video time; the wall clock is held constant);
  • stream, unknown source, FPS 0 and FPS missing (frame_timestamp path).
• has_video_metadata() test in workflows/tests/unit_tests/execution_engine/entities/test_base.py.
• The new tests fail against the old block. Its datetime.fromtimestamp call hits the patched datetime, so they error out rather than show the old output. To show the bug itself, a script with real time ran still images 0.15 s apart with a 0.1 s cooldown: the old block let only the first through ([True, False, False, False]), the fixed one let all four through.

Commands run:

cd workflows && uv sync --locked --extra test && .venv/bin/python -m pytest tests/unit_tests/core_steps/control_flow tests/unit_tests/execution_engine/entities/test_base.py
# 133 passed, 1 skipped
uvx black==26.10.1 <changed files> && uvx isort==5.13.2 <changed files>

Not run: the server integration test tests/workflows/integration_tests/execution/test_workflow_with_rate_limiter.py, which needs a model. It uses no video_reference_image, so its path is the unchanged wall-clock one.

Notes

• Not addressed: the limiter keeps a single last-execution time shared by every video, and doesn't reset when video time goes backwards (a new file or a restart).

[github-actions]

👋 Thanks for the pull request! Here is how automated Claude review works here, so you spend credits (and reviewer time) wisely.

🚧 Right now this is a draft, so automated Claude review is paused — nothing is being spent yet. Mark it Ready for review to trigger it.

Warning

💸 The Claude reviewer bills in credits, not vibes

Automated review spins up a real agent that reads real code and spends real credits on every pass. It is glad to help — but it is not a rubber duck, a linter you poke in a loop, or a substitute for reading the contributing guide. Treat it like an expensive senior reviewer whose time you booked, and show up prepared.

Draft when unsure, Ready when you mean it:

• 🌱 Not sure the PR is in good shape yet? Keep it (or set it back) as a draft — drafts pause review, so you can push and iterate without burning credits on a moving target.
• 💪 Feel strong about the contents? Mark it Ready for review and the reviewer will take a look.

However you get there, arrive prepared:

• 🧱 Bring a SOLID, thorough PR. Point your local agent at our skills/ to tune it to our guidelines first — or, if you are one of those fabled carbon-based contributors, read them yourself. A half-baked diff costs exactly the same to review as a finished one.
• ✅ Resolve every comment before you re-request review. Re-requesting with threads still open means paying twice for the same conversation.
• 🔁 Do not use CI review as an inner loop for a local agent. The reviewer is not a step-by-step debugger — do the unfolding locally and arrive with the answer, not the search.
• 🙋 If something looks off, ask a human. One question to a maintainer is cheaper and faster than three rounds of agent re-review chasing a misread.

Reviews are not free. A draft costs nothing to review; a Ready PR is a promise that it is worth reviewing.

• Prefer to skip automated review entirely? Add the skip-claude-review label.